Kallisto calculates the TPM, which is the "transcript per million". It makes this calculation by working out how many reads come from each transcript, dividing by the length of the gene to caculate reads per kb (FPK). It then divides the FPK for each transcript by the sum of the FPK for all transcripts.
Now, as far as Kallisto is concerned, the transcript doesn't change in length between your knockouts and wildtype (even though it does). If each exon was 100bp long, and you had 100 read(-pairs) per exons in wildtypes, then that would be 800 reads/800bp = 1 read/bp. In the mutant you've got 600 reads, but as far as kallisto is concerned, the transcript is still 80bp long, so you have 600reads/800bp = 0.75 reads/bp. So you would see a reduction.
Now, depending on the struture of your gene, it might not be the case that removing 25% of the exons removes 25% of the length of the transcript. In particular final exons are generally around 1/3-1/2 of the total length of the transcript, meaning that other exons are generally shorter. If the total length of the exons you've removed is short, then you might not see a difference.
However, with todays technologies with 150bp reads, you'll probably find that reads cover more than one exon, particularly if the exons are short. In this case, all reads that overlap the removed exons will fail to map, not just those that map directly to it.
So basically, the answer is complex, and not obvious to work out - you probably should see some reduction, but it might not be that much.
I'm intrigued by the fact you confirmed a reduction in RNA levels using qPCR, as one wouldn't _necessarily_ expect that removal of exons would reduce the expression of a transcript (but probably would lead the production of a non-functional product). However, it could be that your removal of these exons has caused a frameshift early enough in the protein that one of the remaining introns is more than 55nt downstream of the new stop codon, and therefore non-sense mediate decay is triggered.