This has been tested in the Ruby implementation.

Try it online!

This has been tested in the Ruby implementation.

After obfuscating "Hello, World!" in MarioLANG I looked into golfing it a bit. The above is the shortest I have found so far.

As before I started from a Brainfuck solution which sets four cells to the nearest multiple of 10 to the characters He, and space and converted it to MarioLANG. You can then shorten the code a bit by making use of the auxiliary floor in the loop which almost halves the width of the loop. Note that the bottom is only executed one time less than the top, so you don't get exact multiples of the initial counter in all 4 cells any more.

After obfuscating "Hello, World!" in MarioLANG I looked into golfing it a bit. The above is the shortest I have found so far.

As before I started from a Brainfuck solution which sets four cells to the nearest multiple of 10 to the characters He, and space and converted it to MarioLANG. You can then shorten the code a bit by making use of the auxiliary floor in the loop which almost halves the width of the loop. Note that the bottom is only executed one time less than the top, so you don't get exact multiples of the initial counter in all 4 cells any more.

MarioLANG, 259 249 242 240 bytes

+>+>)+)+)+++)++++((((-[!)+.).+++++++..+++.)).(-.((+++++++++++++++.).+++.------.--------.)+
+"+"===================#=================================================================.
 +++!((+++++++++)++++++)<
++=#==================="
++
 !
=#

Finally, I wanted to make use of the wasted space in front of the loop, so I added a bunch of elevators to make use of the vertical space there.

Oh and then there's a little trick at the end of the main row which saves two bytes.

MarioLANG, 259 249 242 240 235 bytes

+>+>)+)+)+++)++++((((-[!)>->.
+"+"===================#+".")
+++!((+++++++++)++++++)<.---+
++=#===================")---.
++((.-(.)).+++..+++++++.<---
 !+======================---
=#>++++++++++++++.).+++.-!>!
  =======================#=#

Finally, I wanted to make use of the wasted space in front of the loop, so I added a bunch of elevators to make use of the vertical space there. And then I realised that I could fold the code after the loop (see previous revision) below the loop to make use of some more vertical space, which saved five more bytes.

Note: This solver assumes that the linear code after the loop is all on the top line, and the above code is that solution folded up. There might be a shorter overall solution by making the solver aware of the folding, because now I get 3 more +s in the first part for free, and the next 4 instructions would cost only 1 byte instead of 2.