More info:

• scd killscd works when in the replug-without-hibernate case, for hibernate, a restart of pcscd is needed (probably not a scd problem).
• when I don't issue any commands to scd while the reader is unplugged, I can re-plug it anywhere (same or different usb port) and scd recognises this as a USABLE->ACTIVE->NOCARD->USABLE transition chain
• when I issue a SERIALNO command while the reader is unplugged, I get "Card Error" back (sitenote: shouldn't that be a ERR_NO_CARD_READER?) and that one sticks until a KILLSCD.
• this is not the case for other commands (tested: GETINFO version).

GPA works fine since it's full of sentinels blocking reentrance into the
reloading code, which I assume is triggered by the scdaemon behaviour
described. In addition, GPA causes the whole stack to wake up periodically to
reload data that didn't change. That's not nice for laptop users and
contributes to Global Warming :)

scd getinfo version

D 2.0.13-svn5059
$ ./src/gpa --version
gpa 0.9.0-svn996