Firstly, is there actually a transfer rate issue here?
If not, do whatever you like :)
However, if there is an issue, the nature of the system will determine the best protocol. A metric that looks at average amount of the actual information you want transmitted per unit time - that is, with all the start/stop, error-checking and addressing bits removed - is what is important.
If your system dictates that you'll be updating every value every time, a valid approach is to do a bulk transfer by ditching the identifiers and relying on the ordering to unpack it.
Alternatively, maybe you wont update every value, every time, but can still opt for the approach just outlined - that is, knowing that you're 'wasting' the redundant info, but still gaining the advantage of not having the addressing bits.
The other extreme is similar to what zoomcat suggests - addressing all transfers - this would be a valid approach for a system with lots of variables that are updated infrequently and/or more randomly.
Also, keep in mind that serial comms has a 'hidden' overhead per transfer - so in general, you'll gain a higher average rate by sending bulk blocks, but at the same time get a higher latency. (google 'latency vs. throughput' ...)
Have a think about what is most important to your system, sometime it'll be clear and the system will determine (force) one approach over the other. Sometimes you're 'lucky' and free to have a harder think about it (heh heh).
My system needed reasonable throughput (within a spec), had a natural latency via other factors that I could work within aaaand the values were updated every period - hence I opted for a bulk transfer of all values with no addressing.
2hrs a day avg, 5days a week avg, 6months so far
2*5* 52/2 = 210 hrs
at 115200 baud
210*60*60*115200 = 87091200000 bits
3 byte packets: 87091200000 / 24 = 3628800000 successful data transmissions so far
Maybe I bodged the math, logic and facts here - but the point remains that the number will be higher than if I had addressed the packets... and, I can't recall ever seeing a failure (the system is such that it should be apparent).
edit, yes I did bodge it! this assumes constant data transfer and no overhead ...
But yeah, YMMV and it's system specific. Good luck to you! :)