An SLA is still an unmaintainable way of looking at the bandwidth planning problem. Companies like Comcast cannot offer an SLA using cumulative data caps: they would have to do instantaneous bandwidth caps.
They would thereby need to divide the available channel capacity by the number of users, and would only be able to guarantee "nearly no bandwidth", even if in practice there is constantly capacity to share.
You instead need to think of bandwidth as a public good that is threatened by tragedy of the commons. You have been given access to a field (to graze cows) for a fixed entrance fee, as have a number of other farmers.
This is simple and works great, because most people own two cows, and the field, over the course of a month, can handle everyone bringing their own cows and grazing them. The price is then adjusted for two cows.
However, as initially the way you do this is to just give the keys to the gate to everyone who paid, the result is going to suck: some people are going to realize that they can graze their friends cows, or their giraffe.
Even worse, some asshat is going to just sit there lighting grass on fire just because it is fun; maybe they even have a reason: they are scientists looking into the properties of grass; it is still messing with the pricing.
I will then claim that it is entirely a fair thing to say "ok, you can't graze more than X amount of grass a month" to make the pricing stable. The alternative, infinite grass for fixed cost, simply doesn't make any sense.
There is, however, a second problem that is analogous to the one that you bring up: that everyone's cows become hungry at the exact same time, and there isn't even really room in the field to graze 100 cows /at once/.
This problem is simply not going to be fixable without firm schedules on who is allowed to be in the field at what times; otherwise you are going to just end up with "we can make no guarantees".
Now, you might argue that "no, Comcast can just build out", but now the field is going to have to be 50x larger than it otherwise would need to be to handle this prime time usage, and the price will also need to be 50x.
Maybe you want to pay 50x to get a reasonable amount of bandwidth during rushes on the system, but honestly, as much as I care about and pay extra for bandwidth, I can't imagine I could afford "what it really costs".
We then need to spend time looking at the "firm schedules" problem, and the solution that bandwidth providers are going to eventually need to realize they must adopt is variable pricing depending on when you access it.
This is already done for other utilities: your power costs more during some hours of the day, as does your cellular telephone service. That bandwidth doesn't cost more is a temporary mistake that will also be corrected.
It must be pointed out, though, that in the "pay by the kilobyte with variable pricing" model, you still don't get an SLA. If everyone decides it is worth it to pay for prime-time-powered air conditioning today, there will be a brown out.
Even so, the result /does/ solve your "demanding fixed payment for variable value" complaint: you simply can't get many kilobytes during the prime time usage, so your kilobyte usage meter will not go up much under congestion.