I've seen those erros too when trying a string inlet on polyphonic subpatches, and haven't found a quick diagnosis or solution.
Working on a fixed point version, a/b test is the v2 patch, gives a nice 40% reduction in cpu power, BUT it blows up a frequencies higher than 12kHz... I'm investigating...
Well, one solution for doubling the cutoff frequency of vcf3 was given in the Full bandwidth filters -thread (which is now closed).
However, and I wanted to discuss that there, it's not that simple. Turns out that the vcf3 oscillates at ca 6kHz, so not even nyquist/2 but nyquist/4, and nyquist/2 is what my quick fix in that thread offered. Therefore it's no surprise that the doubling was very obvious. I did coax it to go to nyquist, but then the filter blows up - this is of course not supposed to happen with the rbj cookbook filters. I discussed it a little with Johannes, but we didn't really conclude. Johannes says there's plenty of room in the internal FLASH, and there's no problem accommodating several more complete filters, as opposed to the wrapper I initially proposed. So a vcf3 with a doubling of the cutoff can be easily had, but there is a question of the coefficients already being calculated to support very clever scaling, so that such a simple modification brings out a somewhat altered filter response.
I think the matter needs a deeper look-into. BTW, I tried my hand at a 24dB ladder, but it blows up at high resonance - a bit irritating as I've successfully implemented it on other systems (not a fault of Axoloti I hasten to add, it's just me not being done with it yet...). I look forward to see Johannes' implementation.
If Johannes and you decide that new filters are on for 1.0.11 (or in the near future) I might look into things again, although there are people better qualified than me here...