I think you are possibly confusing two different issues. Throughout I'll be quoting from this page, which is superb and worth a read.
Firstly we see this:
The whole available frequency band is split into a distinct number of sub-channels.
So we split our entire available frequency up, the page also says that sub-channels may be referred to as "carriers, bins or buckets" if used for data (instead of signalling).
Not all possibly channels are used and not all are used for data. Importantly, as Wouter says, the different frequencies used will not affect speed.
However, if the sub-channel is used for data, it's a "bucket" or "bin", then the following is relevant:
The amount of data bits that can be carried in the bin can vary depending upon the quality of the signal at the particular frequency range for that particular bin.
So if you examine two sub-channels that have been allocated for carrying data you might find that one can carry data at a "faster speed". This means more data at a time rather than "quicker" data that arrives faster.
This is because the signal quality is higher (less noise, less interference, etc.) - all the standard reasons in the black art of RF engineering.
Because the data doesn't arrive "quicker" or "slower", or before or after data sent at the same time, reconstructing it is not a particular challenge. It is the same as reconstructing data if all bins carried the same amount of data per time slot, there is just slightly more data from some bins.