Because it was useless to begin with.
For a more detailed answer the whole picture is needed:
Back in the real old days modems were symmetrical. A V.21 modem split the line into two 300 bps channel, one for each direction. So single direction transfer would use only a half the available bandwidth. This stayed true including the 2400 bps standard (V.22bis). Janus was developed (IIRC) in the 1985-87 time frame, including adaptions for FIDO and so on. For 2400 bps Modems it did improve transfer times. So far a great idea.
Around the same time USR did introduce their proprietary HST modems. HST was based on an asymmetric channel distribution. One direction did get about 9600 bps bandwidth, while the other got only 300 bps. So perfect for single direction transfers. USR also got a huge foothold in the US by applying an aggressive marketing toward BBS operators with substantial discounts, sometimes up to 60% off regular retail. Now, while running Janus on a symmetric line is a great idea, it totally sucks on asymmetric connections. Thus, BBS running a high speed modem were better off utilizing single directional protocols.
Since V.32 was already standardized years before (1984?), using a V.32 modem would have been a better idea, just they were prohibitive expensive (I remember Hayes asking for tier fist V.32 Modem in like 1988 more than 2000 USD, compared to a HST sold below 1000 USD) - and as said before, USR already had a great hold among BBS operators. With the advent of lower priced standard (V.32) modems USR did crank up the HST (16.8 and 19.2) to stay ahead, but ultimately lost. Still, this wasn't until like 1992/3, and many BBS were still on a locking to the HST.
The same time standard 14.4/28.8 modems became available at reasonable prices, the mid 90s, is also when BBS lost it's appeal and major dial in networks or at last the internet took over.
So, yeah, Janus was a good idea, but market forces meant different.
On a more professional side it was never really something to be considered. People connecting to Unix systems or interconnecting Unix systems were mainly using Telebits PEP modems. For one, they not only offered a dynamic channel adjustment, but also protocol spoofing. PEP split the bandwidth into 512 channels each could be assigned to either direction, so the bandwidth was dynamically allocated. Protocol spoofing meant that the modem recognized a transfer using uucp-g or ZModem and handling all protocol issues locally, thus reducing turn around times to zero.