After the molecular markers lane, staining in lanes 1, 3, 5 seems ok, what do you thin?

LAnes 2, 4, 6... are strange and it seems (to me) that the type of sample is different than the type of sample in lanes 1, 3, 5… In principle this seems to be the issue, that the solvent in samples 2, 4, 6… is not suitable for this type of assay.

Is it possible to modify the solvent of these samples (using a PD10 column for instance)?

I should clarify, the samples in all lanes are from the same extraction (and hence in the same solvent) only loaded at different concentrations. I am trying to detect small MW proteins, so not being able to stain the bottoms part of the gel in the latter lanes is the issue at hand

Just to make sure... Were the gels immersed well in the silver stain? Do you use a shaker?  To me it looks as if the lower right part of the gel might have floated up during the staining. Do the gels remain even/flat or do they become "wavy" when they are immersed in the silver stain? If they are wavy, they might have become too warm during the electrophoresis.

You should study your standard protein ladder. Are all the described bands visible? If the last band is the smallest MW marker of that ladder, then your problem is sample migration, not staining. 16.5 % is a high density gel, and it may be that the samples have not penetrated farther than what is shown as stained. Try running the sample for a longer time, or higher voltage (unless its a home-made gel, then just do longer time). I am a big fan of pre-cast Bio-Rad TGX 4-20% gradient gels. They can get very warm without melting or compromising the sample migration. (like, 300 V for 15 min, and then microwave 1 min on HIGH with stain to quicker staining.)

I don't have any experience with silver stain - but I have a lot of experience with SDS-PAGE, and it looks very much like your samples aren't penetrating the gel - it doesn't look like a staining problem. I mean - look at the second to last lane - its a U-shape. Does silver stain 'u-shape'? No. But proteins often run in weird sinusoidal curves, esp if they are 1) overloaded or 2) in a weird solvent that 'plays' with the voltage (urea? Gdn-HCl?, another high Molarity salt?). Try using a precast gel, or a lower % acrylamide gel, try running the gel for longer. Make fresh running buffers CORRECTLY (some people really don't!)

Lastly, try staining with the normal SDS-PAGE dye - this might give you the same pattern and let you know that its your protein migration, not staining, that is negatively affecting your desired outcome. 

Agreeing with these answers , I still wonder whether you are not modifying the concentration of the solvent …. but may be you are loading more amount the sample ? Are you sampling more micro liters in lanes  2, 4, 6 … (more than in 1, 3 and 5)?

If so, I agree with Nicole and also, this is (to me) the main reason for samples not running well, i.e. that the sample is affecting the running. 

I go back to one of my suggestions but in broader terms. I suggest to decrease the amount of "matter" in the lanes. If a crude extract why not doing a pre-purification step and get rid of the heavier proteins? If the protein concentration in the samples is not high (then the problem is not due to the proteins but to the solvent) it would be the the more solvent in lanes 2,4,6…  that gives strange running. In summary get rid of big proteins and/or of any  suspect molecule present  in the solvent.