We present a reanalysis of the He ii Ly \alpha absorption towards the quasars HS~1700+6416 and HE~2347-4342 using new high S / N , optical observations .
An alternative analysis method is applied , which fits the high quality , optical H i data directly to the He ii spectrum .
The results are compared to those inferred from standard line profile analyses .
This new method enables us to derive redshift scales characterizing the fluctuations of the column density ratio \eta .
We find \eta changing smoothly with redshift on typical scales of \Delta z \sim 0.01 - 0.03 corresponding to 8 - 24 h ^ { -1 } \mathrm { Mpc } comoving .
The real length scales of variations of the column density ratio might be even larger , since part of the fluctuations may be caused by noise in the He ii data and by systematic effects due to the applied method .
However , variations on small scales can not be ruled out completely .
For \eta variations on scales of a few Mpc an amplitude of about \pm 1.5 \mathrm { dex } can not be excluded .
The data shows an apparent correlation between low \eta regions and the presence of metal line absorbers , which corresponds to the more general correlation of low \eta and strong H i absorption .
Thermal line broadening is suggested as a probable explanation for this apparent correlation , since both fit methods would severely underestimate \eta for absorbers with N _ { { H \textsc { i } } } \gtrsim 10 ^ { 13 } \mathrm { cm } ^ { -2 } if the line width was dominated by thermal broadening .
Indeed , lines located close to the cut-off of the b ( N ) distribution yield lower column density ratios compared to the whole sample , in particular if high density absorbers are considered .
We argue that the apparent correlation of \eta with the strength of the H i absorption is caused by insufficient consideration of thermal broadened lines by the standard analysis .
As unbiased value of the column density ratio , we find \eta \sim 80 in agreement with previous estimates .