I don't see what difference it makes if the upgrade is an add-on or in the cartridge. It's the same technical upgrade. At least if it's a seperate add-on you're not making people pay for the upgrade over and over again when they shouldn't have to. From a marketing point of view making people pay for it with each cartridge works better.
I say if the cost of the in-cartridge upgrade is nominal than its fine. If, for example, it quadruples the cost of the cartridge than its not fair. With any system that has an audio/video input in the cartridge port you can literally put anything you want on the cartridge.
It certainly makes a practical difference. With a seperate add-on, you obviously have to make sure enough people buy it, which will never be as many people as have the console. That limits the number of cartridges you can produce. At least with in-cartridge chips, you can produce exactly the number you think will sell to the entire console owning base. Even if it makes for a higher-priced cartridge, those costs can be balanced to a degree with the possibility of a much higher production run.
The other advantage of in-cartridge chips is that you're not limited to the technology of the time of the chip's creation. You can keep developing better helper chips. With an add-on, you're limited to what you can do at the time. Sure, you have more room to get it right the first time, but you're still on lock-down going forward. Combined with the other negatives, it seems to me that the use of in-cartridge helper chips is the superior option.
And yes, in theory, you can put an entire system-on-a-chip in some of these cases, and just use the existing console for AV and IO. But in practical terms, we've seen simpler augmentation. Bottom line to me, if I can buy a system on day one and play a game 10 years down the road - helper chip inside or not - I have to say that that's part of the system's core capabilities. I get that these enhancements are not capabilities within the system itself, but for all practical purposes, it's transparent to the end user. That's also why I've always personally favored a practical analysis of system capabilities using existing games rather than theoretically on paper. Theoretical technical prowess means nothing without real evidence of said prowess.
Obviously, with the above in mind, I won't go so far as to say that helper chips count in a raw system-to-system comparison of paper specs. That would be silly. But listing that option as a feature of the system, absolutely. And again, to me, those games count the same as "standard" games when comparing libraries.