Thanks, that explains it. Did a bit more reading in the meantime and hope to have a slightly better understanding by now.
I was actually thinking of something like using a 24-48MHz oscillator at Extal0 and routing it straight through with MCGOUTCLK.
From what I read out of the SDK examples, they are all using the external 32.768kHz as clock source and generating a 47.972352MHz system clock, correct?
Unfortunately, my initial problem still exists: when I am running LPTMR driver example the time is not exactly 1s but about 1.0007s. Which is no real issue on a 1s timebase. But when I switched to 1ms by changing LPTMR_USEC_COUNT to 1000[us], the time base was 1.7ms, which is far out of range.
Looks like I got something wrong with the IRQ rather then system clock?