Is this really matched only to within 10%?
Quick datasheet check:
Indeed, hFe seems well matched between the two, but keep in mind that these curves represent typical characteristics. Only values in the characteristics table are guaranteed by the manufacturer:
Under the specified conditions (Ic=4A Vce=4V) the minimum hFe is 20 and the maximum is 70. The hFe value around 50 from the curve above is right in the middle, so we should kinda expect a gaussian distribution, but they're not telling about its variance. Besidees, hFe varies a lot depending on temperature:
Curves are from another transistor but you get the idea. Since we're talking about power transistors, they will presumably be used in a power amp and thus they will heat, and not necessarily to the same temperature, also temperature will vary, and maybe one transistor is mounted on a cooler part of the heat sink, etc.
In other words, the truth is "between 20 and 70" under specified conditions, which is nothing like this "hFe matched to 10%" thing.
Having similar hFe for both transistors means that both will load the driver circuit (which provides their base current) in the same way and this should reduce some forms of distortion.
This is what you would normally expect from a "complimentary pair": two transistors that the manufacturer will try to make quite similar, but really not enough to be called "matched". Besides, it is not possible to manufacture a NPN and a PNP with identical characteristics.
If you want two transistors to be as matched and identical as possible, the only way is to manufacture them from the same silicon wafer, and have them next to each other so they are made in the exact same conditions. They will have to be of the same type (ie, 2x NPN for example).
Example: SSM2220 or much cheaper DMMT series. In this case the transistors are low power, in the same package to try to keep their temperature as close as possible, and they are either on the same silicon die or from the same silicon wafer with closely controlled matching, verified by testing. A very important bit about these matched transistors is Vbe matching, as this is what determines the input offset voltage if they are used as a long tailed pair.
PS: If you want to build an audio amp, 2955/3055 in class B is not really recommended.
It is easier to match NPN and NPN because they are the same fabrication process and you can make both transistors in the exact same conditions, same wafer, same fabrication run.
However, if they are not made next to each other on the same wafer, stuff like doping, vapor deposition etc is never perfectly uniform across your wafer, and it will change across wafers... so your two "identical" transistors will have lots of dispersion in their characteristics... it's a fact of silicon chip manufacturing.
NPN and PNP use different fabrication processes, with different steps executed in different order, which will depend on different parameters... so matching them is even more difficult.
Thus, 10% is actually not bad at all.
An analogy would be: a machine which makes screws will easily be able to make almost identical screws in the same fabrication lot as long as it uses the same tools and settings, no need to adjust it super accurately. But if you want to make nuts that match the screws exactly, then you have to adjust your nut-making machine to match your screw-making machine very precisely, which is much harder...