The danger of paralleling two decoupling capacitors of very different values doesn't often bite you.
The purpose of a decoupling capacitor is to create a low impedance to ground at all relevant AC frequencies. Relevant for a given circuit is defined like this. At a low enough frequency, the circuit will often tolerate ripple, at a high enough frequency it won't respond to it. Relevant means the frequencies between those limits.
If you have two very different value capacitors, then at some frequency, the smaller capacitor will still be behaving like a capacitor, however the large capacitor will have 'gone inductive', meaning that the residual inductance from its physical length and electrode connections is now dominating its behaviour.
The parallel combination of a capacitor and an inductor forms a parallel resonant circuit, which is high impedance at the resonant frequency. If the Q of the circuit is high enough, then the rail has no bypassing at this frequency.
So, if there is a signal getting onto the rail at this frequency and the circuit is sensitive to this frequency and the capacitors are high enough Q to make a high impedance resonance then you will have problems.
Generally, aluminium electrolytics are so lossy that they will not support a high Q resonance. Often a length of track between a big bulk C at the edge of the board and a chip-local decoupling C is enough to kill any resonance. Often people will use a small resistor or a ferrite bead between the bulk and the local capacitor for filtering, but it will also de-Q any resonance. And then if you have a resonance, if there's no signal at that frequency, you won't notice any effect.
One common situation for hitting a problem is in an RF circuit, where two ceramic decoupling caps of different sizes are put close together. The circuit will often oscillate and find the frequency it can get good feedback along the rail.
The other situation is when decoupling a logic IC, and the particular clock or data signal hits the resonance. It can be quite fun to find data dependent misbehavior, that's there for some frequencies and not others. Imagine the conversation, 'my burst DMA works for byte access, but not for word access, but I can access words OK, what's going on???'