I must admit that I find it counter-intuitive that HLASM would store the double ampersands in the variable, rather than undouble them during the assignment. But well - as long as the results are consistent - I really shouldn't need to care.
As far as I know, it's always been that way. Sometimes such difficulties are due to the fact that SETC expressions involving paired ampersands result in two ampersands -- they aren't paired, while apostrophes are. That is, &X SETC '&&''' is three characters long, not two.
