On my system:

''.unpack('U*')

Produces:

=> [8364]

I would have expected this:

=> [342, 202, 254]

In fact, I could have sworn that things used to work this way... Am I
going crazy? The following seems to confirm that the string is indeed
using a UTF-8 representation internally.

''.collect
=> ["\342\202\254"]

I get exactly the same results whether $KCODE is set to 'NONE' or 'u'.

Cheers,
Greg