retaining majority for 6 months leads to many more than 6 months for the minority to try over taking. because they are many blockheights ahead.
losing advantage/chance as time goes on

for the last 8 years people have debated the whole 51% attack vector that can undo months/years worth of work.
but that those fears are dismissed when you actually calculate what is required for a minority to overtake to then gain blockheight(and chainwork).

if 51% A - 49% B
in the first block
460,000X0 - (event trigger)
then
hashrate vs hashrate B 'could' get height become majority
but chances are A get the next block

460,000X0 - 460,001A1
B then see 460,001 and reject it.
B then see 460,001 and reject it.
blah blah blah. they cant sync

at worse B ban communication to A for B to build their own chain(altcoin)(lets imagine B they reconfigure their node in 10 minutes (quick recode and banning) meaning A gets ahead in that setup period that B is wasting
A: 460,000X0 - 460,001A1 - 460,002A2
B: 460,000X0
by which time B are then a block or 2 behind.
B then start mining and making a block, but ofcourse A is also making blocks too
460,000X0 - 460,001A1 - 460,002A2 - 460,002A3
460,000X0 - 460,001B1

over an average of a day(144 blocks) with a 2% differential. A could gain ~3 blockheight
460,144A - 460,144A
460,141B - 460,141B

over an average of a week(1008 blocks) with a 2% differential. A could gain ~50 blockheight
461,007A - 461,008A
460,957B - 460,958B

so now B chances of overtaking are slimmer.

over an average of 2 weeks(2016 blocks) with a 2% differential. A could gain ~100 blockheight
462,015A - 462016A
461,915B - 461,916B

B have not created 2016 blocks in 2 weeks. so their difficulty drops(meaning they can make blocks with less chainwork)
you may think this makes B have the advantage because now they can make blocks faster. but... here is the clincher

if B accelerated by having a few % lower chainwork needed advantage due to lower difficulty. they are still 100 blocks behind. so aswell as building blocks they have to make up for the loss. meaning
if A just made 2016 blocks in the next 2 weeks. B has to make 2116 just to get even.

and here is the punch line.
chainwork.
if it got to a point of
464,031A - 464,032A
464,031B - 464,032B

464,032B's chainwork will be lower than 464,032A chainwork.
imagine each block cost 100 chainworks per block to solve
over the 4 weeks A has a chainwork of 403200 for block 464034A
over the 4 weeks B has a chainwork of 389088 for block 464034B

meaning if B joined the network to attempt an over throw, it gets orphaned anyway, not due to blocksize, not due to hashpower, not due to blockheight, but due to chainwork.