A) The assumption that science mana output is a linear function of population.
B) The quoted 2% penalty per pop.
One pop is 2% of 50 pops, so when you go from 49 to 50 then you are taking a 2% penalty to increase your output by 2%. Above this point each pop you add has a penalty that's more potent than the bonus it adds; below this point each pop adds more than its penalty is worth.
I´m sorry but that just doesn´t add up.
The base tech cost is 100 science, for this example. The more pops you have, producing the same average amount of science per pop, the easier research gets actually:
Example for 50 pops
Assume you have 50 pops producing 2 science each time unit, adding up to 100.
Due to you having 50 pops, the cost is bumped up by 100%, making it a total 200 to research.
-- Your pops will finish researching it after 2 time units.
Larger Example Base
900 pop, 1800 science, 1900 cost - 1.05x
500 pop, 1000 science, 1100 cost - 1.1x
200 pop, 400 science, 500 cost - 1.25x
100 pops, 200 science, 300 cost - 1.5x
75 pops, 150 science, 250 cost - 1.6x
50 pops, 100 science, 200 cost - 2x
25 pops, 50 science, 150 cost - 3x
The formula is simple (As long as you are above 10 pops) Tech cost = base_cost * (1 + pops * 2 / 100), whilst your Research production = pops * average_science_per_pop
As you see it´s more expensive at the start due to the "+ 1" on tech cost, as our pops is smaller, as pops approaches infinity, with both values converging towards infinity at the same speed, the difference starts to even out, and your most efficient research will be at the largest amount of pops you can possibly have.
( That is not taking into account how efficient you are at building, what ethos/traits and government you have ... generally speaking, there are science governments that can focus down on tech, whilst there are militarist governments that focus on conquest. The tech govs will probably be ahead due to their internal focus and resource allocation - and cards they draw )