Yes, you are taking one flaw with the XP system and replacing it with another.
If you have 20 Kobolds (CR 1/8), they all by themselves generate a CR 3-4 encounter roughly. Those weak monsters significantly contribute to the difficulty of the encounter. But multiplying the XP of the CR 5 monster by the 21 monsters is equally stupid.
Hence things like
F$&% CR, There’s a Better Way (Part 2) -- where you can deconstruct the system so that it generates nearly the same results when you have X monsters of CR Y compared to one monster of CR Z, but also doesn't break down when you have 20 Kobolds and a CR 5.
Now you can do more math than the angryGM did and work out that a CR 5 monster and 20 CR 1/8 monsters comes to an encounter level of around 8 (this is napkin math from memory; actual result will be +/-1) for a 4 person party. Which disagrees with the XP math.
But if you just add up XP you get XP equal to a CR 6 encounter. If you use the multiplier, you get a ridiculous result, because most of the XP comes from the CR 5 which then upscales way too high. And you get that funny jump once you switch from no multiplier to multiplier.
For about 80% accuracy, do this:
1) Map CR 1/8, 1/4 and 1/2 to 1/3, 1/2 and 2/3. (footnote 1)
2) Map CR above 20 to (Original - 20)*2.5 + 20. So a CR 30 is mapped to 50. (footnote 2)
Now add up the CR you get.
Sum of PC levels/5 or higher: Easy
Sum of PC levels/4 or higher: Medium
Sum of PC levels/3 or higher: Hard
Sum of PC levels/2.5 or higher: Deadly
Party of 3 level 5 PCs? Sum is 15, so Easy is 3+, Medium is 3.75+, Hard is 5+, Deadly is 6+.
Party of 5 level 5 PCs? Sum is 25, so easy is 5+, Medium is 6.25+, Hard is 8.3+, Deadly is 10+.
---
When rebuilding existing encounters for a party of X, what you care about is the scale factor more than anything. So you just apply above, then scale for party size.
---
Footnote 1: Doing this for sub-1 is important, becuase you often have large masses of these. And error adds up. There is also error from 1-5, but it is smaller (it should go 1, 1.8, 2.6, 3.2, 4 or so, but that is within the 80% accuracy bound, so skip it).
Footnote 2: Basically, above-20 is 2.5 "real" CR per CR.