The repeated characters like that is what I would expect to see if you tried doing an 8 bit text decode on a 16 bit script* (common enough failure in ROM hacking). For it to happen on retail is very odd indeed as most text engines take no real cue from the base system, though not impossible something like that is happening. Is the baseline system modded (be if Japanese to play English or English to play Japanese) at all? If it is forcing something to behave as English then it could be set up in such a mode by that, which is presumably why the alternative loader might be better equipped (although still trying to read 8 bits in at least the initial game info screen by the looks of that). You would then have to tackle what is going wrong there and indeed whether you can do anything -- if you have an alternative of load it in Swiss then that might be the solution if say a mod chip is rather more annoying to update/tweak.
*Japanese Kanji total thousands of characters, even for the reduced sets, so you can't get away with 8 bit (2^8=256) encodings whereas Roman character based stuff is not so constrained; 26 characters, 52 for both cases, 10 numbers, some punctuation and a few more for the non English squigglies can still afford Ancient Greek, Cyrillic and a bunch of other things with room to spare in 256. Because computers work best in multiples of 8 bits then 16 is the next obvious choice and what logographic languages tend to have to use. All that said Japanese in a lot of simplistic (which means games as they are for kids) might stick to the simpler Hiragana and Katakana (collectively kana) which number and in some ways behave rather more in line with the western alphabets and thus the padding to get it to 16 bits reflects in the repeated characters.