TM, itself, should likely solve the problem thusly:
- launch the process and start emitting output into buffer as it is
produced
start buffer-n-spew mode after ##K worth of output
suspend the inferior and bring up a "continue/continue-for-1MB-
more/quit" after, say, a megabyte of output produced
That would solve the problem without trying to turn the TextMate editing environment into a log monitor.
Considering that probably TextMate never will be able to handle such big amounts of data as quick as Xcode for the mentioned reason, I conclude that this proposal is reasonable.
But besides the question of proper policy regarding sub processes and big amounts of data, there's still the problem that TextMate in fact quits ungracefully (that means without asking whether so save any data) when running "yes" from a document, aborting the process and waiting until this to happen.