...why not connect your keyboard to the L3t rear panel Line In jacks? These inputs are more compatible with a keyboard stereo output than the L3t side panel channel mixer.
That's (typically) excellent advice from silverhead, gp4leigh...
Excuse me for now prattling on, but I do so in the hope that some may find it helpful to have a little more background, ...in terms of explaining just WHY it's our feeling that this may solve the problem.
At first glance, it would seem that silverhead's only talking about the difference in levels and impedances between what emerges from "line" and "mic" outputs, but (unlike less "forgiving" older-generation equipment) StageScape and StageSource "XL combi" inputs have been very carefully crafted by Line 6 to effortlessly accommodate all of that, albeit with the occasional use of that "pad switch" for higher output devices, ...exactly as you've done.
So what else may be going on ...and why ?
(I thought you'd never ask.)
First off, the "key" to our "diagnostic thinking".
(Are we all sitting comfortably. Good. Uncle Ron will now tell us a story.)
The digitally connected (L6 Link) part of our system (the L3m) was only subject to encoded information that already had any "rough edges" (that may have bedevilled our original analog inputs) "knocked off" by the very process of those input signals having been "encoded" into L6 Link data...
...So it kept working ...and whatever the problem is, (or was) that behaviour indicates to us that it almost certainly only exists (or existed) in "The Land Of Analog", a misty and magical kingdom of (often impenetrable) fog and confusion.
Now, for the rest.
As we all know, "keyboards" are legendary beasts that have long been the subjects of myths and epics, but one of the very few well-documented, demonstrable and scientifically proven FACTS about them is that many have the power to emit sounds of almost supernatural dimensions in parts of the spectrum that we mere humans cannot even hear.
I'm talking about both sub and supersonic oscillation, which can be present at a level that's high enough easily "trip" a system's "self protection" circuitry even though the offending "preset" or "voice" from the keyboard was only perceived to be at a "low" level.
It's a fair bet, then, that when our...
...Second song was a louder song and the L3t sprang into life and was fine all night...
...a different "preset" or "voice" was used, ...one which did NOT contain the offending sub or supersonic component that we're suspecting.
Why might the "Line" input/s be any better ?
(Thanks for asking.)
The "XL combi" inputs that may be at the heart of this issue are specifically engineered to accept the entire bandwidth of whatever may come from the very finest microphones, from the lowest harmonics of kick drum and double bass all the way up to the highest, such as those that may emanate from the "ting" of a triangle or bell tree.
"Line" inputs, however, are nearly always configured to pass ONLY the stuff we humans can actually hear, on the basis that any harmonic "degradation" will have been accounted for in whatever "sampling" or "mixing" process that occurred prior to them being used.
So no sub or supersonic "demons" from them.
(And they all lived happily ever after. ...Aah, ...we hope. )