There's an interesting phenomenon here. As previously mentioned, my mind has partially assimilated to losing the s/sh distinction. I had a chinese tutor who merged those two sounds, and it stopped bothering me. The same tutor also merged n/l and r/y. I didn't assimilate to those; hearing n where the word I knew used l confused me every single time.
What's relevant is that she didn't merge n/l in her english (which was not fluent). That suggests to me that she might have been able retrospectively to identify which she had just used for a mandarin word like 能 (neng). I agree that that depends on having been trained to recognize the distinction, although I find it slightly odd that she wasn't trained to make it in Mandarin. But, grand summary, I suspect a Japanese speaker could be trained to recognize the distinction even in Japanese.
More generally, I don't think the phenomenon of not distinguishing two different sounds (your Japanese speaker) is the same as the phenomenon of fantasizing that you're distinguishing two identical sounds (your english speaker).
> The same way that you perceive "a" and "A" to be the same grapheme despite seeing different shapes, Japanese speakers consider [r] and [l] to be the same phoneme /R/ despite hearing different sounds.
I know that "a" and "A" are the same letter, but I'm easily capable of distinguishing them on demand. In fact, I'm likely to get upset when one is used where I think the other should be. I don't think that mirrors the r/l situation in Japanese. Similarly, I don't vary freely between A and a (and, owing to personal aesthetics, never write α unless I have to write cursive); I follow a fairly clear set of rules that determines one or the other. I know that situation does not reflect the r/l situation in Japanese, although I don't know that it doesn't reflect the situation of any given speaker.
Given a hypothetical Japanese speaker who's capable of producing r and l allophones, if you asked them to repeat a sentence including "are" realized as "ale", down to mimicking the accent, how likely do you think they would be to mimic that realization of [l]?
But it also holds true that it gets easier to learn new languages, if you do it a lot. I've heard a lot of people say that it gets easier after the fifth language. I'm only up to four (Norwegian, English, Japanese and French - with a basic understanding of other European "dialects" - German, Spanish, Italian -- and of course "Scandinavian dialects" of Swedish and Danish) -- and I can imagine that if I learned one more "truly distinctive" language, like Finnish, or Mandarin -- I'd start to know more of the "possible ways" we use sounds -- and it would therefore be easier to choose "the right subset" for a new language.
In addition to mimicking accents, singing can be a very useful way to train pronunciation.
The same way that you might write A or a or even α to write the letter we pronounce as "aye", a Japanese speaker might choose to pronounce [r] or [l].
A Japanese speaker won't be able to tell that they're saying [r] in an English word that has an /l/. The same way that an English speaker might SWEAR they say writer and rider differently until you record them and they have no clue which one they said when.