@ Repentant Evildoer,
We used, or considered using, just such methods as this. We routinely deleted or hid our own components, or made them load as printer drivers. We randomly shuffled the innards of our loader so anti-virus programs wouldn't recognize it as it was being downloaded...
Hmm you were one of the "Guns for Hire" I used to talk about at that time when much of the security industry and academic researchers were trying to portray such malware activities as being "uber script kidies" doing such things for "ego food"rather than for monetary gain.
I could be petulant and go "Nah Nah told you so!" to those industry "pundits / gurus" but to be quite honest I'd rather just smile politely and carry on making my predictions (and hopefully living long enough ;-) to see if I call them right or not.
And tucked away in the article Bruce links to is a little snippit that is not in the MS posting. Which is the bit about checking for user activity at the keyboard etc.
That is it's checking for a real life user machine and many would think not a server. But that would be a very limited view point, what is not mentioned is that this behaviour also acts as a Honey pot/net filter, thus making it less likely that the malware will get snagged.
Some years ago I directed my mind to how to avoid Honey pot/net systems and user activity or lack there of was my first thought, but I considered it "to obvious".
However another thought that I discussed on this blog was how to remotly detect such systems in as camoflaged a way as possible.
The realisation you need is not to look for user activity which can be faked with scripts if the admin is smart enough but something else.
Which is where the old "follow the money" argument comes into play.
All Honey trap systems use resources which are expensive in various dimensions such as system purchase, power consumption, environmental cost (ie physical space, air con etc), maintanence and insurance costs etc. To reduce costs nearly all honey trap systems use Virtual Machine technology and this can be seen as an Achilles Heel of Honey Traps.
So the question then becomes not how to detect users but how to detect virtual machines.
After a little thought you realise you are looking for a distinquisher that shows two or more virtual machines share the same hardware from a remote location...
Well there are several ways but my criterion was how to camoflage what you are doing from the Honey Trap admin. Who is for obvious reasons paying very close attention to their logs but in all probability also recording every packet that hits the network in ways that cannot be remotely detected. So the method must not only appear as someting the admin expects in their logs and will thus ignore but also not contain any differentiator like a precious new zero day exploit etc.
There are several ways this can be done but most leak some information about what you are doing if the person looking at captured data is smart enough.
Thus the probe you use must at the very least follow the "Duck Principle" (If it looks like a duck, waddles like a duck and quacks like a duck, why would you assume it's a goose). So you need to find a suitable duck, and I decided to use a nice noisy "brain dead script kiddie attack" that does a simple port test or equivalent that looks just like netcat etc in a simple script. Obviously what ever method you chose has to get above the access bar set by the Honey Trap admin.
The real difference is that the probe is carefully timed and the responces from the VM's are likewise carefully timed. This is because VM's all share the same system clock and thus any drift in it can be seen in the VM responses across the network due to packet timing information. Thus if you see a bunch of machines that all have the same clock drift the chances are they are all on the same base hardware and are thus VM's and thus best avoided if you don't want your latest valuable zero day to be caught and analysed.
I actually emailed one or two people in the HoneyNet project to warn them to be alert for such a probing attack or do what would be required to decouple the VMs sufficiently well from the system clock that a network timing probe attack would be prohibitive in time. Of those emailed the response was shall we say compleatly underwhelming.
Perhaps now they realise malware writers are looking to detect Honey Trap machines they will take the issue a little more seriously...
 For those who want to make their own predictions two pieces of advise and a filter. Firstly "follow the money" ultimatly it is "the route of all evil" which motivates people to tred that path to supposed riches which is the base of the American Dream. Secondly study history in a broad context as has been observed "people who do not learn from history have condemed themselves to re-live it" or more appropriatly new tech alows old tricks new places to be performed. So find an old trick and mentaly re-work it for the new environment and if you can see how money can be made from it, there's a high degree of certainty that somebody will actually do it. When is the question and that's where you apply the filter, it will happen most probably when the easier money has been nearly all consumed and the smarter people see it's time to move to a new trick (ie the principle of the low hanging fruit).