Oneline tests running:
10 T=MILLIS(1): FOR I=1 TO 10000: A=5: NEXT: PRINT MILLIS(1)-T

Explanation:
 
In this loop the interpreter processes the following per iteration:
5 token reads (VARIABLE A, =, NUMBER 5, :, NEXT)
2 heap searches (for A and I)
1 assignment
6+numsize byte reads
As a rough over the thumb estimate 50% of the runtime goes into
the two heap searches and 50% in token and byte read operations.
This may vary a lot.


Times in milliseconds

Floating point results (BASICFULL, LONG NAMES, NO NETWORK): 

Boardtype	ms	loops/ms
Arduino GIGA:	82	117
TTGO ESP32:	258	39 
WROOM ESP32:	371	27
Rasp RP2040:	588	17 
Arduino DUE:	589	17
Wifi R4:	629	16
ESP8266:	1178	8
XMC1100: 	1442    7	
MEGA256:	2198	5

BASIC can do between 5 and 117 loops per millisecond. The typical 
timescale for one loop of 32bit boards is 20 microseconds. 
Everything in this range can be done with BASIC. 

Integer results (BASICSIMPLE, SHORT NAMES):
UNO R3: 	1800	5.5.
LEONARDO:	1574 	6.3 (Tinybasic, static vars)


Floating point results (BASICFULL, LONG NAMES, NO NETWORK, 64bit FLOAT): 
Arduino GIGA:	85
ESP8266:	1479	

The GIGA board has a double precision hardware FPU and can do process
the additions and compares of the loop as fast as with 32 bits.