Some tips in order to all process:

• avoid to recreate browser
• do not require inside function
• check to not to serialize the creation
• use pool components in order to expose service (puppeteer-pool, puppeteer-page-pool, browser-pool)
• as underground system is the browser, check differents options in order to tune the browser (headless) that is using for load pages

Pointing to creation

• check if browsing the page using no headless browser is getting delay (maybe is the page that have problems to load because some script or innecesary content)
• check the pdf creation using browser , maybe the content loaded is the problem: extra size, too complicate converting process. Remember that there are some tweaks converting images and other content.