If any non-Unicode connection has some statements prepared and appropriate SQL text contains non-ASCII characters and the MON$ tables are queried in an Unicode connection, this error is thrown by any v2.1 version.
In v2.1.3 RC1, the situation is even worse: error may happen even if the MON$ tables are queried using the same attachment charset, provided that it's not Unicode. The patch to fix the regression and revert to the pre-v2.1.3 behavior is a trivial one-liner, but a complete fix is somewhat more difficult.
I've been hit by this on a number of occasions. What I do (and recommend my users to do) is use the same charset on the monitoring attachment and on other client attachments.
Unfortunately this is not always possible or desirable.