deletions | additions       diff --git a/experim.tex b/experim.tex  index 1b0ddcd..5d8b22a 100644  --- a/experim.tex  +++ b/experim.tex 
...
\end{small}  \end{table}  Unfortunately, we are unable to compute direct performance metrics for the solution by Lameed and Hendren since its source code has not been released. Figures in their paper~\cite{lameed2013feval} show that for these benchmarks the very same MATLAB programs  the speed-up of the OSR-based approach is on average within $30.1\%$ of the speed-up of hand-coded optimization (ranging from $9.2\%$ to $73.9\%$); for the JIT-based approach, the average grows to $84.7\%$ (ranging from $75.7\%$ to $96.5\%$). Our optimization technique yields speed-ups that are very close to the upper bound given from by-hand optimization; in the {\em  worst case - {\tt case} ({\tt  odeRK4} benchmark - benchmark),  we observe a $94.1\%$ when the optimized code is generated on-the-fly, on the fly,  which becomes $97.5\%$ when a cached version is available. Compared to their OSR-based approach, the compensation entry block is a key driver of improved performance, as the benefits from a better type-specialized whole function body outweigh those from performing a direct call using boxed arguments and return values in place of the original \feval.