Having 10uF instead of 1uF is 10 times larger than datasheet specifies.
Having 10uF instead of 0.1uF is 100 times larger than datasheet specifies.

There is still no one here who have involved bit error rates or anything.

But still no linearity involved in anyones argument.

All that has been noted is that smaller deviations from nominal values tends to mean less than large deviations from nominal values - components are normally designed to have safety margins for non-critical component values.

But safety margins are introduced to support ageing, manufacturing tolerances, temperature coefficients etc. So chip while manufacturers must make sure there are safety margins, they don't have any reason to introduce safety margins of ten or 100 times.

If you do care about statistics, you will find that for a random circuit, a larger deviation from rated intervals will increase the probability of misfunction or component failures. That goes for deviations in temperatures, supply voltages, resistance values, capacitances, magnetic fields, ...

So it can be seen as worse to have 100 times too large capacitor than to have a 10 times too large capacitor. And it is possible that both circuits does work if you test them. But how to test? Test for full supply range? Test for full temperature range? Test for time from power-on until stable voltages? Test for fault percentage after one billion power cycles? Test for temperature increase from short-circuiting outputs? Test that Vcc never gets higher than the max value from the datasheet during startup?

In the case of a MAX232A with 0.1uF capacitors, it's likely that Maxim has designed the chip to work well with 1uF capacitors just to make sure that it can directly replace a MAX232 even if someone forgets to change the capacitors. But what incentives would Maxim have for supporting 10uF capacitors, which could potentially increase the initial load on the output transistors during a short-circuit?

In the end, the probability of problems is larger with 10uF capacitors with a MAX232A than with a MAX232 (ignoring that MAX232 in SO-16W socket has 0.1uF). But no linearity exists. And the big problem is that many times it isn't possible to test if it is ok to use too large capacitors, because it isn't possible to know the critical parameters. We don't have any Spice files and other design documentation for the internals of the chip.

And ending up with a situation where we can't just conect an instrument and measure how well a component value works, all we can do is switch to the empirical knowledge that there is a general, non-linear, rule that the further away from documented component values, the worse it generally is.

Looking at the Maxim datasheets for MAX222/242 that have a shutdown feature (designed for 0.1uF capacitors), the datahseet specifies max 250us startup time with the rated 0.1uF capacitors. Looking at the plotted graph, it can be deduced that the voltages have reached about 90% after about 150us with 0.1uF capacitors, but requires about 1ms with 1uF capacitors. And with 1uF it takes about 3ms until the chip has full voltage swing. With 10uF instead, a quick-booting processor might send the first characters with out-of-spec voltage swing.