I count about 140 or so if we stick just to words that are still written down somewhere and were written by jan Sonja. I think I may be 1 or 2 off from your count depending on pu (which is merely reserved), and noka (which seems like it might be "archaic", but the tea leaves published by jan Sonja are hazy) In the corpus, use of the words other than the 125 most "official looking" words is pretty negligible in the corpus.
Even if jan Sonja took a break from work and issued an edict blessing all 140 or so words as official (and she hasn't), then toki pona is still a small language, with an incredibly small number of morphemes.
I think the biggest accounting scandal is not talking about the lexeme count (side stepping what a lexeme *really* i)s, if lexemes include all thing set phrases, compound phrases, idioms, "unofficial phrases", then toki pona has at least 3000+, just counting the entries in the dictionary jan Kipo compiled (the one in MS-Excel format). If you include the proper modifiers, then tp probably easily has another 1000 proper modifiers than a fluent toki pona speaker would need to memorize. I haven't attempted to, but it's on my to do list to mechanically parse the entire corpus into distinct noun phrases, which I suspect would run into the 10's of thousands. Some large percent of those are the sort of things that a person would memorize because some phrases are just so hard to make up or hard to expect people to understand on the fly.
The killer problem for small language design is that the # of meanings we'd like to express is so large and so many meanings are fundamentally hard to transparently describe in a short noun or verb phrase that large scale lexicalization is likely to creep into any small language that is being used as a language.