Basically the quote you have shown was written by someone who either didn't understand what they were talking about, or oversimplified it.
The bandwidth of the cable is a result of the resistance of the cable and the fact that it is capacitive. These act like an R-C low pass filter limiting the bandwidth. Additionally the distributed inductance and capacitance of the cable are frequency dependent so have a more complex affect on the bandwidth.
The "insertion loss" which is a measure of the gain of the cable is dependant on both frequency, but also on length. The longer the cable, the more lossy it is.
Here is one example of the insertion loss of a CAT5e cable, this for a 100m length:
Image Source
Here we see that at 100MHz, the loss is actually more than 20dB - a lot more than the 3dB (power gain) or 6dB (voltage gain) point that would be used to specify the bandwidth of a first order low-pass filter as the quote implies.
Further up the Wikipedia page more accurately explains where the 100MHz figure comes from:
The specification for category 5 cable was defined in ANSI/TIA/EIA-568-A, with clarification in TSB-95. These documents specify performance characteristics and test requirements for frequencies up to 100 MHz
Basically, the requirements in terms of insertion loss that must be met for a cable to be classified as Cat5 are only specified up to 100MHz. Beyond this point the specifications of the cable are undefined by the standard - though manufacturers may well provide data at the higher frequencies.
In terms of how the Ethernet speed relates to frequency, that is a touch more complex. The raw data is encoded using various schemes which are described in brief in the wiki page here. For 100BASE-TX, several encoding schemes are used which result in a maximum fundamental frequency of 31.25MHz through the cable itself.