Stuff tried (relative to CVS head):

(1) use normal apr_pool_create() in worker for ptrans

    still segfault (dunno where; no dump on Linux)

(2) move back to apr_pools.c/1.125

    still segfault

(3) switch to debug pool code (and fix the mutex initialization):

    still segfault

(4) use normal apr_pool_create() for ptrans, define APR_POOL_DEBUG

    still segfault

(5) use normal apr_pool_create() for ptrans

    still segfault

(6) efence on Solaris

    segfault down in efence

(7) efence on Linux

    segfault somewhere, no dump, can't see segfault in gdb