Adobe Target one variant with significantly less traffic than the other: explain sample mismatch

We are running an AB test to test updated Headline copy on a landing page. We are testing 4 variants (including control) and have manually set the allocation to 25% to each. I understand that audiences will never exactly match the allocated manual distribution, however in our latest AB test one variant is seeing a significant misallocated. 

Some background:

• Allocation manually set to 25%
• This is an AB test for updated headline copy, no technical or component changes and no difference in how the variants were set up.
• This is the ONLY test active on the site, there is no collision with other tests
• There is no audience or segment applied: all users are eligible
• The gap is present for the duration of the experiment (so its not a one time blip)

I am struggling to determine what would cause this and if invalidates the results of the test. 

Has anyone experienced something similar? How can we avoid in the future or prevent?