I'm not sure if this is the schematic
@falco_femoralis is referring to, but I suspect this is a representative example:
PedalPCB Snozzberry (One Control Strawberry Red). This schematic doesn't explicitly label all the VREFs, but you can see there is a local, dedicated voltage divider and capacitor everywhere the bias voltage is needed (I counted four, one per opamp).
As for
why... there are different reasons. In some cases, it's almost certainly not necessary, and likely just a copy-paste propagation of an earlier error, or at least copy-paste of a circuit fragment taken out of context. I can't think of an example off the top of my head, but I'm pretty sure I've seen Chuck consolidate superfluous bias voltages into one when he gives a circuit his treatment.
As for a legit reason: depending on how VREF is used, in some cases, it might be sourcing or sinking non-trivial amounts of current, in which case isolating the current-heavy part might help keep it cleaner/more stable for other parts of the circuit. In the Snozzberry, at least for IC1.2, I suspect it was done so that the VREF impedance on the non-inverting input exactly matches the impedance of the inverting input. But I suspect the design would also work essentially the same with a single, low-impedance VREF (say 10k+10k divider, 47uF cap), and localized impedance-matching resistors.
