Hot standby read slaves exceed max delay on WAL segment. Replication lag.

This is on a production 9.0.15 install with 1 master, 5 hot standby read
only slaves.

We've been trying to look into a situation where we're seeing that hot
standby read slaves are receiving WAL segments, but are exceeding
max_standby_archive_delay (60s) and max_standby_streaming_delay (60s) and
not applying changes. The slave will get the first segment and hang (we've
seen up to 30m before removing the slave from our read pool to catch up)
and get further and further behind the master. Furthermore, we have seen
that after a slave has caught up, putting it back into the read pool will
mean will almost immediately start to see this happen again. It acts as if
we had max_standby_* set to -1.

I'm just looking for some ideas, hints, or suggestions as to what might be
going on here or what we might be doing wrong.

Thanks,
Shaun