Suppose you have a working copy at revision 866, and the repos is now at
revision 1000 (i.e., a typical mass-commit scenario). If you try to
update from the top of your working copy, it will fail. Neon complains
"unable to read chunked response". If you go down to a subdir and update
one file at a time, or a small subdir at a time, the updates will succeed.
If you do enough of these, then you'll eventually be able to finish off
the update by updating from a higher directory, because now there will
be much less svndiff data coming down the pike (at least, that's our
conjecture as to the reason, but the symptoms point to a firm correlation
between amount of diff data and likelihood of update failure).
This is, just barely, not an M3 bug. People can do their updates in
pieces for a while, if necessary. But it will be painful, hence the
P1 priority.
Blocker SVN-481 optimization: don't open two TCP sessions for updates
