Indeed, besides pushing a new bitstream he also implemented difficulty >1 support, wow!
Yeah, because I had a similar unexpected-downtime experience with EclipseMC -- they recently started rejecting two-low-difficulty shares.
Unfortunately since the X-Reject-Reason headers aren't standardized the safest thing to do is assume that when a share is rejected that work ought to be thrown out. This wrecks the hashrate.
Well, the reasons
are standardized as part of
BIP 22.
Obviously it's not necessarily a rule for the now-obsolete getwork protocol, but there's no reason
not to use the same strings either.
But more importantly: if your miner had supported the target - which
has been standard since getwork was created - you wouldn't have these rejects at all.