I had the same issue when sending too many commands to the binding or more precisely the binding to the Amazon servers.
My only solution was a group with every echo as a member or use GroupItem Methods to build something like a dynamic group.
I don´t know why but this works…
I also noticed that sending commands too fast will end in bad request errors.
For example: Triggering a rule with only one TTS command multiple times one after the other.
This shouldn´t be a problem.
I already called this address from other machines than openHAB where the binding is running.