Benchmarking is the act of running a standardized test or set of tests on a system (like a computer) to assess its performance. These tests are supposed to provide a consistent measure of how well the system performs under specific conditions.
From the current machine implies using the machine itself to measure its own performance. Herein lies the paradox:

    Self-Reference Problem: When you benchmark from the current machine, you're essentially asking the machine to judge its own performance. This self-assessment can introduce biases or inaccuracies because the machine's current state (e.g., load from other processes, hardware conditions) might affect the benchmarking results.
    Baseline Comparison: Benchmarks are most useful when they allow for comparison between different systems or the same system under different conditions. If you're only benchmarking from one machine, you miss out on comparative data unless you're comparing against previous benchmarks of the same machine, which still could be skewed by the machine's current state.
    Hardware and Software Variability: Even on the same machine, results can vary due to software updates, background processes, or hardware degradation over time. This variability can make "current machine" benchmarking less reliable for drawing conclusions about performance over time.
    Oxymoronic Nature: The term "oxymoron" here highlights the contradiction or irony in expecting an unbiased, accurate performance measure from the very system that's being measured. It's akin to saying "the sound of silence" in benchmarking terms - the act of measuring might alter what is being measured, especially in real-time performance scenarios.

Control for Variables: If done with careful controls (like ensuring no other significant processes are running, using stable benchmarks, and understanding the machine's state), you can still get meaningful data. 
Immediate Feedback: For developers or system administrators, benchmarking on the current machine can provide immediate insights into performance under specific configurations or after certain changes.
Historical Comparison: Over time, with consistent methodology, you can track performance degradation or improvements.

Component-level measurements being published as system performance
Various confusing metrics used by vendors
Lack of credible performance information and agreement among competing vendors3

Operating system efficiency
Compiler performance
Libraries
Algorithms
Application software3