The idea of mirroring is obviously that if one side of the mirror fails, the other should take over. In an ideal world, the sides should also work in tandem to increase read performance when both sides of the mirror are available.
That said, if one side of the mirror fails, then all the in-flight reads to the failed device will fail, possibly after a delay. This is normal and expected: a command was sent to a device that suddenly is no longer there and able to respond to the command, which will result in an error condition of some kind. The kernel will most likely log these failures to give the administrator a heads-up that "something bad just happened". The system may be configured to output these important kernel events to the console.
The litmus test for any mirroring solution is whether these errors actually propagate to the userspace layer, resulting in user applications receiving I/O errors (or worse, invalid data). If a mirror setup is working properly, as long as the other side of the mirror works properly, userspace applications should be unaffected save for the fact that the read took a little longer than usual and the system spat out some diagnostics about I/O errors occuring on the now-unavailable device. Neither of these should appreciably impact well-behaved userspace software.
If the userspace processes (rather than just the Btrfs code in the kernel) saw I/O errors as a result of your experiment, and you can reproduce the behavior at least reasonably consistently, then you may have come across a bug in the Btrfs code. In that case, you may want to file a bug report. Especially given that this is Debian, I would suggest first filing the bug report in Debian's bug tracking system and let them escalate it to the kernel developers if they feel that is warranted. Make sure to include as much pertient detail as possible, including the exact commands you are running, exact versions of everything involved, the exact text of the error messages, an exact description of your storage setup, and anything else you can think of that may help in tracking down the problem.