A fine research question. A lot of the questions from
https://blog.torproject.org/blog/lifecycle-of-a-new-relay
apply here too.
One theory is that it's a result of variance -- they're both relatively
small relays, so one stream that uses lots of bandwidth and randomly
happens to pick this relay could be enough to skew a few bandwidth tests,
pushing you down for a while.
In theory it should recover, since both relays are publishing similar
underlying self-measured bandwidths:
https://exonerator.torproject.org/serverdesc?desc-id=825feea5d643a228e121d0fbfdecf47acff4870f
https://exonerator.torproject.org/serverdesc?desc-id=ec726fe88dbc60d18b3402fff2bace3564292546
But it might take quite some time to adjust, since the bandwidth
authorities don't measure very often.
--Roger