I've got some ceramic 100uF capacitors (1210 100uF X5R 6,3V 20% CL32A107MQVNNNE by Samsung). They came from a local supplier quite a bit cheaper than they sell at the major outlets like Mouser - this is a low-selling item at the local supplier and apparently had not been hit by the current ceramic-craze, yet...

Using one of them instead of the 3x10uF on the L476-DISCO's RTC yielded almost 10 minutes of RTC run. This again is better than what the charge equation would indicate (100uF * (3V - 1.7V) / 300nA is cca 7 minutes), but it roughly follows the expectations.

When the voltage drops below some 1.7V where the 32kHz oscillator stops working, the RTC's consumption significantly drops, so it retains the "stopped time" for tens of minutes - and when power is restored, it continues where it left off - that's why it's easy to know the "maximum endurance".

Of course this is still not what we want - even if the leakage is much lower than indicated (DS says 100 MOhm*uF with the usual method of "Rated Voltage, 60~120 sec." - which is orders of magnitude off the real, non-DA-confused leakage), it's still impractical to pack more than a dozen of them, and that buys no more than two hours of running the 'L476 clock. Clearly, a lower-consumption RTC is needed - waiting for the mentioned low-consumption module to arrive.

And the dielectric absorption thing is HUGE... measurement running, stay tuned!