First, regarding the difference in counts - I'm sure it's related to different settings of unpacked archives for the scans you're running.
The boot-time scan is "more thorough" in that it scans early during the boot process (so the potential malware might not have started yet, i.e. couldn't have hidden itself), and that it uses a raw disk access (which again might prevent some hiding to succeed). But it certainly isn't able to unpack more archives than the usual scanner engine (slightly on contrary, actually).
As for the sleep... interesting.
You might be basically right - or at least that's the only explanation I can think of right now. To prevent the computer from sleeping during the scan (which was reported a few times in the past), I added a piece of code that tells windows "don't sleep while this process is running". I'm not actually sure if it had any effect, because I don't think we could reproduce the original behavior... but that was the only thing to try.
Now, maybe... the system doesn't activate the sleep during the scan, because of this, but the timers are counting - and as soon as the scanner terminates, they turn the value accumulated during the scan into an "effect".
[If this theory is true, however, the behavior should depend on how long the boot-time scan has been running. I.e. if you let it scan just one folder instead of all your disks, to make the scan very short (not exceeding the sleep timeout) - it shouldn't happen afterwards.]
I can try to reset the counters just before the boot-time scanner exists (in one of the following virus database updates)... whether it will have any effect, no idea.
